Fix the issue of GPU hang when encoding picture with one macroblock width

Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Reported-by: Xiang Haihao <haihao.xiang@intel.com>

10 files changed, 66 insertions, 38 deletions

diff --git a/src/gen6_mfc_common.c b/src/gen6_mfc_common.c
index 1796003..bb5d360 100644
--- a/src/gen6_mfc_common.c
+++ b/src/gen6_mfc_common.c
@@ -889,6 +889,7 @@ gen7_vme_walker_fill_vme_batchbuffer(VADriverContextP ctx,
 	int num_mb = pSliceParameter->num_macroblocks;
 	unsigned int mb_intra_ub, score_dep;
 	int x_outer, y_outer, x_inner, y_inner;
+	int xtemp_outer = 0;
 
 	x_outer = first_mb % mb_width;
 	y_outer = first_mb / mb_width;
@@ -932,7 +933,10 @@ gen7_vme_walker_fill_vme_batchbuffer(VADriverContextP ctx,
 	    x_outer += 1;
 	}
 
-	x_outer = mb_width - 2;
+	xtemp_outer = mb_width - 2;
+	if (xtemp_outer < 0)
+		xtemp_outer = 0;
+	x_outer = xtemp_outer;
 	y_outer = first_mb / mb_width;
 	temp = 0;
 	for (;!loop_in_bounds(x_outer, y_outer, first_mb, num_mb, mb_width, mb_height); ) { 
@@ -972,14 +976,11 @@ gen7_vme_walker_fill_vme_batchbuffer(VADriverContextP ctx,
 		x_inner -= 2;
 		y_inner += 1;
 	    }
-	    temp++;
-	    if (temp == 2) {
+	    x_outer++;
+	    if (x_outer >= mb_width) {
 		y_outer += 1;
-		temp = 0;
-		x_outer = mb_width - 2;
-	    } else {
-		x_outer++;
-	    }	
+		x_outer = xtemp_outer;
+	    }		
 	}
     }
 
diff --git a/src/shaders/vme/inter_bframe_haswell.asm b/src/shaders/vme/inter_bframe_haswell.asm
index 331841f..9002b5f 100644
--- a/src/shaders/vme/inter_bframe_haswell.asm
+++ b/src/shaders/vme/inter_bframe_haswell.asm
@@ -316,6 +316,9 @@ mov  (2)  mbc_result.20<1>:w		0:w	{align1};
 
 jmpi   (1)    mb_mvp_start;
 mbd_start:
+mov  (8) mb_msg0.0<1>:ud	0:ud		{align1};
+and.z.f0.0 (1)		null:uw	input_mb_intra_ub<0,1,0>:ub	INTRA_PRED_AVAIL_FLAG_D:uw   {align1};
+(f0.0)  jmpi (1)	mb_mvp_start;
 mov  (1) mbc_result.0<1>:d	MB_AVAIL		{align1};	
 mov  (2) tmp_reg0.0<1>:UW	orig_xy_ub<2,2,1>:UB	{align1};
 add  (2) tmp_reg0.0<1>:w	tmp_reg0.0<2,2,1>:w	-1:w	{align1};
diff --git a/src/shaders/vme/inter_bframe_haswell.g75b b/src/shaders/vme/inter_bframe_haswell.g75b
index 02759d9..d769fe1 100644
--- a/src/shaders/vme/inter_bframe_haswell.g75b
+++ b/src/shaders/vme/inter_bframe_haswell.g75b
@@ -37,7 +37,7 @@
    { 0x00600001, 0x2ac00061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2a800061, 0x00000000, 0x00000000 },
    { 0x01000005, 0x20002d28, 0x000000a6, 0x00040004 },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x00000cb0 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000ce0 },
    { 0x00600001, 0x2ae00061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2b000061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2b200061, 0x00000000, 0x00000000 },
@@ -64,7 +64,7 @@
    { 0x00000001, 0x2fa40021, 0x00000b80, 0x00000000 },
    { 0x00000001, 0x2fa80061, 0x00000000, 0x00000001 },
    { 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
-   { 0x00000020, 0x34001c00, 0x00001400, 0x000013a0 },
+   { 0x00000020, 0x34001c00, 0x00001400, 0x000013d0 },
    { 0x00000001, 0x2aa00129, 0x00000fe4, 0x00000000 },
    { 0x01000010, 0x20002d28, 0x00000aa0, 0x00000000 },
    { 0x00010001, 0x2af001e9, 0x00000000, 0x00010001 },
@@ -80,7 +80,7 @@
    { 0x01000005, 0x20002e28, 0x000000a5, 0x00100010 },
    { 0x00210001, 0x2b1401ed, 0x00000000, 0xffffffff },
    { 0x00210001, 0x2b3401ed, 0x00000000, 0xffffffff },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x00000630 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000660 },
    { 0x00000001, 0x2b0000e5, 0x00000000, 0x00000001 },
    { 0x00200001, 0x24000229, 0x004500a0, 0x00000000 },
    { 0x00000040, 0x24023dad, 0x00000402, 0xffffffff },
@@ -100,7 +100,7 @@
    { 0x00000001, 0x2fa40021, 0x00000b80, 0x00000000 },
    { 0x00000001, 0x2fa80061, 0x00000000, 0x00000002 },
    { 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
-   { 0x00000020, 0x34001c00, 0x00001400, 0x00001160 },
+   { 0x00000020, 0x34001c00, 0x00001400, 0x00001190 },
    { 0x00000001, 0x2aa00129, 0x00000fe4, 0x00000000 },
    { 0x01000010, 0x20002d28, 0x00000aa0, 0x00000000 },
    { 0x00010001, 0x2b1001e9, 0x00000000, 0x00010001 },
@@ -126,7 +126,7 @@
    { 0x0a800031, 0x2b601ca1, 0x00000b40, 0x02280303 },
    { 0x05000010, 0x2000252c, 0x00000b70, 0x00000b88 },
    { 0x00210001, 0x2b3401ed, 0x00000000, 0xffffffff },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x00000350 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000380 },
    { 0x00000040, 0x2b480c21, 0x00000b48, 0x00000003 },
    { 0x0a800031, 0x2ba01ca1, 0x00000b40, 0x02480403 },
    { 0x00200001, 0x2b240021, 0x00450bf0, 0x00000000 },
@@ -135,19 +135,22 @@
    { 0x00000001, 0x2fa40021, 0x00000b80, 0x00000000 },
    { 0x00000001, 0x2fa80061, 0x00000000, 0x00000002 },
    { 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
-   { 0x00000020, 0x34001c00, 0x00001400, 0x00000f30 },
+   { 0x00000020, 0x34001c00, 0x00001400, 0x00000f60 },
    { 0x00000001, 0x2aa00129, 0x00000fe4, 0x00000000 },
    { 0x01000010, 0x20002d28, 0x00000aa0, 0x00000000 },
    { 0x00010001, 0x2b3001e9, 0x00000000, 0x00010001 },
    { 0x00010001, 0x2b3401ed, 0x00000000, 0x00000000 },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x00000270 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x000002a0 },
    { 0x01000010, 0x20002d28, 0x00000aa0, 0x00010001 },
    { 0x00010001, 0x2b3201e9, 0x00000000, 0x00010001 },
    { 0x00010001, 0x2b3601ed, 0x00000000, 0x00000000 },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x00000230 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000260 },
    { 0x00200001, 0x2b3001e9, 0x00000000, 0x00010001 },
    { 0x00200001, 0x2b3401ed, 0x00000000, 0x00000000 },
-   { 0x00000020, 0x34001c00, 0x00001400, 0x00000200 },
+   { 0x00000020, 0x34001c00, 0x00001400, 0x00000230 },
+   { 0x00600001, 0x2b400061, 0x00000000, 0x00000000 },
+   { 0x01000005, 0x20002e28, 0x000000a5, 0x00040004 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000200 },
    { 0x00000001, 0x2b2000e5, 0x00000000, 0x00000001 },
    { 0x00200001, 0x24000229, 0x004500a0, 0x00000000 },
    { 0x00200040, 0x24003dad, 0x00450400, 0xffffffff },
diff --git a/src/shaders/vme/inter_bframe_ivb.asm b/src/shaders/vme/inter_bframe_ivb.asm
index c8d9c16..490467c 100644
--- a/src/shaders/vme/inter_bframe_ivb.asm
+++ b/src/shaders/vme/inter_bframe_ivb.asm
@@ -305,6 +305,9 @@ mov  (2)  mbc_result.20<1>:w		0:w	{align1};
 
 jmpi   (1)    mb_mvp_start;
 mbd_start:
+mov  (8) mb_msg0.0<1>:ud	0:ud		{align1};
+and.z.f0.0 (1)		null:uw	input_mb_intra_ub<0,1,0>:ub	INTRA_PRED_AVAIL_FLAG_D:uw   {align1};
+(f0.0)  jmpi (1)	mb_mvp_start;
 mov  (1) mbc_result.0<1>:d	MB_AVAIL		{align1};	
 mov  (2) tmp_reg0.0<1>:UW	orig_xy_ub<2,2,1>:UB	{align1};
 add  (2) tmp_reg0.0<1>:w	tmp_reg0.0<2,2,1>:w	-1:w	{align1};
diff --git a/src/shaders/vme/inter_bframe_ivb.g7b b/src/shaders/vme/inter_bframe_ivb.g7b
index ec967ea..6ab9059 100644
--- a/src/shaders/vme/inter_bframe_ivb.g7b
+++ b/src/shaders/vme/inter_bframe_ivb.g7b
@@ -24,7 +24,7 @@
    { 0x00600001, 0x2ac00061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2a800061, 0x00000000, 0x00000000 },
    { 0x01000005, 0x20002d28, 0x000000a6, 0x00040004 },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x000001a4 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x000001aa },
    { 0x00600001, 0x2ae00061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2b000061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2b200061, 0x00000000, 0x00000000 },
@@ -53,7 +53,7 @@
    { 0x00000001, 0x2fa40021, 0x00000b80, 0x00000000 },
    { 0x00000001, 0x2fa80061, 0x00000000, 0x00000001 },
    { 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
-   { 0x00000020, 0x34001c00, 0x00001400, 0x0000023e },
+   { 0x00000020, 0x34001c00, 0x00001400, 0x00000244 },
    { 0x00000001, 0x2aa00129, 0x00000fe4, 0x00000000 },
    { 0x01000010, 0x20002d28, 0x00000aa0, 0x00000000 },
    { 0x00010001, 0x2af001e9, 0x00000000, 0x00010001 },
@@ -69,7 +69,7 @@
    { 0x01000005, 0x20002e28, 0x000000a5, 0x00100010 },
    { 0x00210001, 0x2b1401ed, 0x00000000, 0xffffffff },
    { 0x00210001, 0x2b3401ed, 0x00000000, 0xffffffff },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x000000d0 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x000000d6 },
    { 0x00000001, 0x2b0000e5, 0x00000000, 0x00000001 },
    { 0x00200001, 0x24000229, 0x004500a0, 0x00000000 },
    { 0x00000040, 0x24023dad, 0x00000402, 0xffffffff },
@@ -91,7 +91,7 @@
    { 0x00000001, 0x2fa40021, 0x00000b80, 0x00000000 },
    { 0x00000001, 0x2fa80061, 0x00000000, 0x00000002 },
    { 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
-   { 0x00000020, 0x34001c00, 0x00001400, 0x000001f2 },
+   { 0x00000020, 0x34001c00, 0x00001400, 0x000001f8 },
    { 0x00000001, 0x2aa00129, 0x00000fe4, 0x00000000 },
    { 0x01000010, 0x20002d28, 0x00000aa0, 0x00000000 },
    { 0x00010001, 0x2b1001e9, 0x00000000, 0x00010001 },
@@ -118,7 +118,7 @@
    { 0x0a800031, 0x2b801ca1, 0x00000b40, 0x02180003 },
    { 0x01000005, 0x20000c20, 0x00000b80, 0x00002000 },
    { 0x00310001, 0x2b3401ed, 0x00000000, 0xffffffff },
-   { 0x00110020, 0x34001c00, 0x00001400, 0x0000006e },
+   { 0x00110020, 0x34001c00, 0x00001400, 0x00000074 },
    { 0x00000001, 0x2b480021, 0x00000b68, 0x00000000 },
    { 0x0a800031, 0x2ba01ca1, 0x00000b40, 0x02480403 },
    { 0x00200001, 0x2b240021, 0x00450bf0, 0x00000000 },
@@ -127,19 +127,22 @@
    { 0x00000001, 0x2fa40021, 0x00000b80, 0x00000000 },
    { 0x00000001, 0x2fa80061, 0x00000000, 0x00000002 },
    { 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
-   { 0x00000020, 0x34001c00, 0x00001400, 0x000001aa },
+   { 0x00000020, 0x34001c00, 0x00001400, 0x000001b0 },
    { 0x00000001, 0x2aa00129, 0x00000fe4, 0x00000000 },
    { 0x01000010, 0x20002d28, 0x00000aa0, 0x00000000 },
    { 0x00010001, 0x2b3001e9, 0x00000000, 0x00010001 },
    { 0x00010001, 0x2b3401ed, 0x00000000, 0x00000000 },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x00000052 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000058 },
    { 0x01000010, 0x20002d28, 0x00000aa0, 0x00010001 },
    { 0x00010001, 0x2b3201e9, 0x00000000, 0x00010001 },
    { 0x00010001, 0x2b3601ed, 0x00000000, 0x00000000 },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x0000004a },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000050 },
    { 0x00200001, 0x2b3001e9, 0x00000000, 0x00010001 },
    { 0x00200001, 0x2b3401ed, 0x00000000, 0x00000000 },
-   { 0x00000020, 0x34001c00, 0x00001400, 0x00000044 },
+   { 0x00000020, 0x34001c00, 0x00001400, 0x0000004a },
+   { 0x00600001, 0x2b400061, 0x00000000, 0x00000000 },
+   { 0x01000005, 0x20002e28, 0x000000a5, 0x00040004 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000044 },
    { 0x00000001, 0x2b2000e5, 0x00000000, 0x00000001 },
    { 0x00200001, 0x24000229, 0x004500a0, 0x00000000 },
    { 0x00200040, 0x24003dad, 0x00450400, 0xffffffff },
diff --git a/src/shaders/vme/inter_frame_haswell.asm b/src/shaders/vme/inter_frame_haswell.asm
index 79d9a70..fd7b8fa 100644
--- a/src/shaders/vme/inter_frame_haswell.asm
+++ b/src/shaders/vme/inter_frame_haswell.asm
@@ -266,6 +266,9 @@ mov	   (1)		mbc_result.16<1>:w		MB_PRED_FLAG		{align1};
 
 jmpi   (1)    mb_mvp_start;
 mbd_start:
+mov  (8) mb_msg0.0<1>:ud	0:ud		{align1};
+and.z.f0.0 (1)		null:uw	input_mb_intra_ub<0,1,0>:ub	INTRA_PRED_AVAIL_FLAG_D:uw   {align1};
+(f0.0)  jmpi (1)	mb_mvp_start;
 mov  (1) mbc_result.0<1>:d	MB_AVAIL		{align1};	
 mov  (2) tmp_reg0.0<1>:UW	orig_xy_ub<2,2,1>:UB	{align1};
 add  (2) tmp_reg0.0<1>:w	tmp_reg0.0<2,2,1>:w	-1:w	{align1};
diff --git a/src/shaders/vme/inter_frame_haswell.g75b b/src/shaders/vme/inter_frame_haswell.g75b
index 36d9419..e6f17c4 100644
--- a/src/shaders/vme/inter_frame_haswell.g75b
+++ b/src/shaders/vme/inter_frame_haswell.g75b
@@ -36,7 +36,7 @@
    { 0x00600001, 0x2ac00061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2a800061, 0x00000000, 0x00000000 },
    { 0x01000005, 0x20002d28, 0x000000a6, 0x00040004 },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x00000710 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000740 },
    { 0x00600001, 0x2ae00061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2b000061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2b200061, 0x00000000, 0x00000000 },
@@ -63,7 +63,7 @@
    { 0x01000005, 0x20002e28, 0x000000a5, 0x00100010 },
    { 0x00210001, 0x2b1401ed, 0x00000000, 0xffffffff },
    { 0x00210001, 0x2b3401ed, 0x00000000, 0xffffffff },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x00000320 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000350 },
    { 0x00000001, 0x2b0000e5, 0x00000000, 0x00000001 },
    { 0x00200001, 0x24000229, 0x004500a0, 0x00000000 },
    { 0x00000040, 0x24023dad, 0x00000402, 0xffffffff },
@@ -93,12 +93,15 @@
    { 0x0a800031, 0x2b601ca1, 0x00000b40, 0x02280303 },
    { 0x05000010, 0x2000252c, 0x00000b70, 0x00000b88 },
    { 0x00210001, 0x2b3401ed, 0x00000000, 0xffffffff },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x00000140 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000170 },
    { 0x00000040, 0x2b480c21, 0x00000b48, 0x00000003 },
    { 0x0a800031, 0x2ba01ca1, 0x00000b40, 0x02480403 },
    { 0x00200001, 0x2b240021, 0x00450bf0, 0x00000000 },
    { 0x00000001, 0x2b3001ed, 0x00000000, 0x00010001 },
-   { 0x00000020, 0x34001c00, 0x00001400, 0x000000f0 },
+   { 0x00000020, 0x34001c00, 0x00001400, 0x00000120 },
+   { 0x00600001, 0x2b400061, 0x00000000, 0x00000000 },
+   { 0x01000005, 0x20002e28, 0x000000a5, 0x00040004 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x000000f0 },
    { 0x00000001, 0x2b2000e5, 0x00000000, 0x00000001 },
    { 0x00200001, 0x24000229, 0x004500a0, 0x00000000 },
    { 0x00200040, 0x24003dad, 0x00450400, 0xffffffff },
diff --git a/src/shaders/vme/inter_frame_ivb.asm b/src/shaders/vme/inter_frame_ivb.asm
index 33af8a5..d33b146 100644
--- a/src/shaders/vme/inter_frame_ivb.asm
+++ b/src/shaders/vme/inter_frame_ivb.asm
@@ -257,6 +257,9 @@ mov	   (1)		mbc_result.16<1>:w		MB_PRED_FLAG		{align1};
 
 jmpi   (1)    mb_mvp_start;
 mbd_start:
+mov  (8) mb_msg0.0<1>:ud	0:ud		{align1};
+and.z.f0.0 (1)		null:uw	input_mb_intra_ub<0,1,0>:ub	INTRA_PRED_AVAIL_FLAG_D:uw   {align1};
+(f0.0)  jmpi (1)	mb_mvp_start;
 mov  (1) mbc_result.0<1>:d	MB_AVAIL		{align1};	
 mov  (2) tmp_reg0.0<1>:UW	orig_xy_ub<2,2,1>:UB	{align1};
 add  (2) tmp_reg0.0<1>:w	tmp_reg0.0<2,2,1>:w	-1:w	{align1};
diff --git a/src/shaders/vme/inter_frame_ivb.g7b b/src/shaders/vme/inter_frame_ivb.g7b
index bfe3e5a..22f5563 100644
--- a/src/shaders/vme/inter_frame_ivb.g7b
+++ b/src/shaders/vme/inter_frame_ivb.g7b
@@ -24,7 +24,7 @@
    { 0x00600001, 0x2ac00061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2a800061, 0x00000000, 0x00000000 },
    { 0x01000005, 0x20002d28, 0x000000a6, 0x00040004 },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x000000f2 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x000000f8 },
    { 0x00600001, 0x2ae00061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2b000061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2b200061, 0x00000000, 0x00000000 },
@@ -53,7 +53,7 @@
    { 0x01000005, 0x20002e28, 0x000000a5, 0x00100010 },
    { 0x00210001, 0x2b1401ed, 0x00000000, 0xffffffff },
    { 0x00210001, 0x2b3401ed, 0x00000000, 0xffffffff },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x00000070 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000076 },
    { 0x00000001, 0x2b0000e5, 0x00000000, 0x00000001 },
    { 0x00200001, 0x24000229, 0x004500a0, 0x00000000 },
    { 0x00000040, 0x24023dad, 0x00000402, 0xffffffff },
@@ -87,12 +87,15 @@
    { 0x0a800031, 0x2b801ca1, 0x00000b40, 0x02180003 },
    { 0x01000005, 0x20000c20, 0x00000b80, 0x00002000 },
    { 0x00310001, 0x2b3401ed, 0x00000000, 0xffffffff },
-   { 0x00110020, 0x34001c00, 0x00001400, 0x0000002c },
+   { 0x00110020, 0x34001c00, 0x00001400, 0x00000032 },
    { 0x00000001, 0x2b480021, 0x00000b68, 0x00000000 },
    { 0x0a800031, 0x2ba01ca1, 0x00000b40, 0x02480403 },
    { 0x00200001, 0x2b240021, 0x00450bf0, 0x00000000 },
    { 0x00000001, 0x2b3001ed, 0x00000000, 0x00010001 },
-   { 0x00000020, 0x34001c00, 0x00001400, 0x00000022 },
+   { 0x00000020, 0x34001c00, 0x00001400, 0x00000028 },
+   { 0x00600001, 0x2b400061, 0x00000000, 0x00000000 },
+   { 0x01000005, 0x20002e28, 0x000000a5, 0x00040004 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000022 },
    { 0x00000001, 0x2b2000e5, 0x00000000, 0x00000001 },
    { 0x00200001, 0x24000229, 0x004500a0, 0x00000000 },
    { 0x00200040, 0x24003dad, 0x00450400, 0xffffffff },
diff --git a/src/shaders/vme/mpeg2_inter_frame_haswell.g75b b/src/shaders/vme/mpeg2_inter_frame_haswell.g75b
index 1d5b276..5c76728 100644
--- a/src/shaders/vme/mpeg2_inter_frame_haswell.g75b
+++ b/src/shaders/vme/mpeg2_inter_frame_haswell.g75b
@@ -36,7 +36,7 @@
    { 0x00600001, 0x2ac00061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2a800061, 0x00000000, 0x00000000 },
    { 0x01000005, 0x20002d28, 0x000000a6, 0x00040004 },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x00000710 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000740 },
    { 0x00600001, 0x2ae00061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2b000061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2b200061, 0x00000000, 0x00000000 },
@@ -63,7 +63,7 @@
    { 0x01000005, 0x20002e28, 0x000000a5, 0x00100010 },
    { 0x00210001, 0x2b1401ed, 0x00000000, 0xffffffff },
    { 0x00210001, 0x2b3401ed, 0x00000000, 0xffffffff },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x00000320 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000350 },
    { 0x00000001, 0x2b0000e5, 0x00000000, 0x00000001 },
    { 0x00200001, 0x24000229, 0x004500a0, 0x00000000 },
    { 0x00000040, 0x24023dad, 0x00000402, 0xffffffff },
@@ -93,12 +93,15 @@
    { 0x0a800031, 0x2b601ca1, 0x00000b40, 0x02280303 },
    { 0x05000010, 0x2000252c, 0x00000b70, 0x00000b88 },
    { 0x00210001, 0x2b3401ed, 0x00000000, 0xffffffff },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x00000140 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000170 },
    { 0x00000040, 0x2b480c21, 0x00000b48, 0x00000003 },
    { 0x0a800031, 0x2ba01ca1, 0x00000b40, 0x02480403 },
    { 0x00200001, 0x2b240021, 0x00450bf0, 0x00000000 },
    { 0x00000001, 0x2b3001ed, 0x00000000, 0x00010001 },
-   { 0x00000020, 0x34001c00, 0x00001400, 0x000000f0 },
+   { 0x00000020, 0x34001c00, 0x00001400, 0x00000120 },
+   { 0x00600001, 0x2b400061, 0x00000000, 0x00000000 },
+   { 0x01000005, 0x20002e28, 0x000000a5, 0x00040004 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x000000f0 },
    { 0x00000001, 0x2b2000e5, 0x00000000, 0x00000001 },
    { 0x00200001, 0x24000229, 0x004500a0, 0x00000000 },
    { 0x00200040, 0x24003dad, 0x00450400, 0xffffffff },